Computer predicts cancer chances

Click to follow
The Independent Online
TRIALS OF a new "supercomputer", designed to function like the human brain, have shown it is better than doctors at predicting the survival chances of cancer patients and the progress of their disease.

Although the technology is still in its infancy, the research has significant implications for the quality of life of people with cancer and could help reduce NHS spending on unnecessary treatments.

More accurate predictions on the progress of the disease, the likelihood of a tumour recurring and a patient's life expectancy will help cut the need for early surgery, as well as pinpointing the type of treatment needed.

The new study looks at the effectiveness of artificial neural networks - simplified models of the interconnecting neurons in the brain which are trained to find patterns in clinical data - in predicting the course of bladder cancer.

According to Raouf Naguib, Professor of biomedical computing at Coventry University, it is the first time that this technology has been directly compared with consultants' prognoses for the development of cancer.

The Coventry research, carried out with Kilian Mellon, senior lecturer in urological surgery at Newcastle University, and Professor Freddie Hamdey, head of urology at Sheffield University, looked at 212 retrospective case studies of men and women with bladder cancer.

It compared human and computer predictions, based on 15 "markers", including gender, age, smoking habits, information about the tumour and other factors associated with the development of cancer.

When measured against what actually happened, the computers were found to be 82 per cent accurate in assessing the survival rates for patients with advanced cancer, compared to 65 per cent for the urologists over 12 months.

For cases of "intermediate" bladder cancer, which can behave like a superficial tumour or gravitate to a more advanced stage, the computer was 82 per cent correct - twice as accurate as the urologists.

The doctors' prediction rates improved for patients suffering the early stages of bladder cancer but the computer was still more effective. "Predicting the development of cancer is a very intuitive process, with doctors relying on their accumulated experience to come to a conclusion," said Professor Naguib.

"Artificial neural networks, which make associations between a patient's clinical history and social habits, can help doctors come to more accurate conclusions, although they should not be used in isolation.

"Their use could lead to big savings for the NHS by giving a clearer indication of when and when not to operate. It will also eventually mean that doctors can prescribe aggressive treatment for patients who are not expected to live for long and minimum treatment to preserve the quality of life for patients whose cancer is less severe."

Bladder cancer is more than twice as common in men as in women and around 5,400 people die of the disease each year in England and Wales. Around 40 per cent of people diagnosed as having advanced bladder cancer survive more than five years.

However, Professor Naguib cautioned that it would take time to develop the system on a broad scale because, although the computer reaches a decision in seconds, it take months to train and a large amount of information needs to be input.

According to Kilian Mellon, the technology could be particularly useful for patients diagnosed with intermediate bladder cancer, because it would help doctors decide whether to begin aggressive early treatment.

"It should improve on our predictions of what is likely to happen and, as a consequence, means we can better target the people who need treatment," he said.

Lesley Walker, head of science information at the Cancer Research Campaign, welcomed the study, saying it was a part of a general trend, including genetic profiling, of trying to predict how patients will respond to treatment.

The researchers want to measure the system's accuracy over five years, following the condition of patients who have been newly diagnosed with cancer.